Longitudinal Cognitive Development Assessment

We conducted linear growth models (Singer & Willett, 2003 ) to address our RQ1. More specifically, we examined whether children’s assessment scores, including Raven, Corsi, mental rotation, and Panamath tasks, changed across three time points over six months. Given the small sample size, we only used unconditional growth models without covariates. HLM 7.03 software program (Raudenbush & Bryk, 2002 ; Raudenbush & Congdon, 2021 ) was used for conducting linear growth models. The growth model is defined as follows: $$\mathrm{Level}-1(\mathrm{intra}-\mathrm{individual}\mathrm{level}):Y_{\mathit{ti}}={\pi }_{0i}+{\pi }_{1i}Time+e_{\mathit{ti}}$$ $$\mathrm{Level}-2(\mathrm{inter}-\mathrm{individual}\mathrm{level}):{\pi }_{0i}={\beta }_{00}+r_{0i};{\pi }_{1i}={\beta }_{10}+r_{1i}$$ where $Y_{\mathit{ti}}$ denotes the outcome at time t for child i; ${\pi }_{0i}$ is the intercept of the growth trajectory for child i; ${\pi }_{1i}$ is the growth rate for child i; time denotes the indicator of data collection waves (Time = 1, 2, 3); ${\beta }_{00}$ is the average of level-1 intercepts; ${\beta }_{10}$ is the average of level-1 growth rates; $e_{\mathit{ti}}$ is the level-1 residuals; and $r_{0i}$ and $r_{1i}$ are level-2 residuals for the intercept and growth rate, separately.
Thematic analysis (Braun & Clarke, 2013 ) was used for analyzing the interview data, which intended to address our RQ2. We followed a six-step process: familiarization, coding, generating themes, reviewing themes, defining and naming themes, and writing up. The audio interview data were transcribed to text first and then were analyzed by NVivo 12 software program (QSR International Pty Ltd., 2020 ). For the interview conducted in Mandarin, the Chinese transcript was first translated into English by a transcription software (https://otter.ai/); two RAs independently checked and corrected errors in the translation and then went over all the translations with a senior researcher, who had English translation training, to discuss the consistency and accuracy of the translations. If there were any conflicts, the senior researcher resolved the issue with two RAs together.
Eye-tracking data were analyzed to address our RQ3, which were coded by iMotion software (https://imotions.com/eye-tracking/) and analyzed via data visualization using ggplot2 R package (Wickham, 2016 ). Given the small sample size from this subset of data, we did not conduct any statistical analysis. We used line chart to visualize the eye-tracking measures over time and compared all the measures between two locations.